ANGULAR TREEMAPS - Information Visualisation: Techniques, Usability and Evaluation - page 101

Fig. 5.12 presents a comparison of four different treemap layout
results, including Slice-and-Dice Treemaps, Squarified Treemaps, Space-
Optimised Tree, and Angular Treemaps, using the same data set with
approximately 2400 vertices. The evaluation of computational time was
conducted for the same rectangular container with 1000×1000 pixels by
using various data sets. Table 5.1 provides the computational time results
of our Angular Treemaps algorithm compared to other available partitioning
algorithms for different datasets with 272 vertices to 650,000 vertices. The
running time is not necessarily linear to the size of the data but is also
dependent on the complexity of the data and the tessellation process. Table
5.1 indicates that the running time of our algorithms is competitively close
to the three alternative treemaps. It provides the opportunity for Angular
Treemaps to be adopted into real time applications.

Number of
Vertices

Slice &
Dice

Squarified SO-Tree Angular

272 2 4 7 80

2400 7 21 24 120

16,600 27 138 156 220

43,200 31 38 92 234

81,000 49 61 155 343

122,000 70 87 225 186

650,000 319 2903 1010 1505

Table 5.1. The computational time (in milliseconds) of Slice-and-Dice Treemaps,
Squarified Treemaps, and Space-Optimised Tree in comparison with our Angular
Treemaps algorithm on various data sets using the same rectangular container.

Usability study

The purpose of a usability study is to evaluate the hypothesis that Angular
Treemaps have a positive impact in locating and identifying targets of
common visual data analysis tasks. For this usability study, we compared
Angular Treemaps with the vertical-horizontal treemap technique (Į =
90º). Both treemap techniques used the divide-and-conquer approach in
the partitioning process. A preliminarily usability study was conducted
with a control group of eight subjects for the application. All subjects
indicated that they had experience with visual data analysis. The
experiments were performed with 28 datasets and 112 scenarios. The
datasets were categorized by four different types of datasets; each type had
seven alternative datasets, with increasing number of hierarchical level and